Nvidia AI Server Prices Set to Jump More Than 15% in 2027

Server manufacturers supplying Microsoft, Google, and Oracle have told customers to expect price increases exceeding 15% on many systems built around Nvidia's AI chips. The higher costs are tied to servers shipping in early 2027, giving hyperscalers roughly a year's notice to adjust budgets and procurement plans.

The culprit isn't Nvidia's silicon itself but the memory surrounding it. Counterpoint Research found that server DRAM prices roughly doubled in the first quarter of 2026, with DRAM, NAND, and HBM all posting quarter-over-quarter increases of 80% to 90%.

That spike is now working its way through the supply chain into finished server pricing. Because AI accelerators pair with large volumes of high-bandwidth memory, any tightening in memory supply hits AI server bills of materials disproportionately hard compared with conventional compute infrastructure.

The affected hardware spans both current and next-generation Nvidia platforms. Systems built on Grace Blackwell, Nvidia's current flagship architecture for large-scale training and inference, are expected to see the increases, alongside servers designed around Vera Rubin, the company's upcoming platform.

For the hyperscalers footing the bill, a 15% jump on AI server fleets that already run into the tens of billions of dollars represents a meaningful new line item. Microsoft, Google, and Oracle have each committed to aggressive capital expenditure plans to keep pace with AI compute demand, and tighter memory markets could force a choice between absorbing the cost, passing it on to cloud customers, or slowing the pace of deployment.

The timing also matters for how the industry reads memory markets going forward. A near-doubling of DRAM and HBM prices in a single quarter is an unusually sharp move, and it suggests memory suppliers are struggling to keep pace with demand from AI infrastructure buildouts even as chipmakers ramp accelerator production.

Whether the 15% figure holds as a ceiling or turns out to be a floor will depend largely on how memory supply responds over the next year. For now, the notification to server customers gives data center operators concrete numbers to plan around, months before the affected hardware actually ships.